Gemma 4 and Android skills in Android Studio
Short answer: Android Studio Quail 4, stable since September 1, 2026, can run Gemma 4 as a local model: pick Gemma in the agent’s model selector or go to Settings › Tools › AI › Model Providers › Gemma, choose a model, and Android Studio downloads it and runs it with a bundled inference engine. It works fully offline, your code stays on your machine, and there’s no token quota. The smallest models run with 12GB of RAM; Google says 32GB or more works best. Quail 4 also bundles 23 Android skills that the agent applies automatically.
Local Gemma 4 versus a Gemini upgrade
From the Android Developers Blog, September 1, 2026.
| Option | How | What you get |
|---|---|---|
| Gemma 4 local | Model selector › Gemma, or Settings › Tools › AI › Model Providers › Gemma | Offline agent with tool calling; code stays local; no quota |
| Default Gemini | Built in | No cost; capabilities adjusted dynamically by Google |
| API key | Google AI Studio key, or Anthropic or OpenAI keys | Latest models as soon as available |
| Google AI Pro or Ultra | Sign in with your Google account | Premium capacity and higher rate limits |
| Gemini Enterprise | Sign in with your organization | Google Cloud privacy and security; rolling out in Canary |
Setting up Gemma 4
Open the agent’s model selector, choose Gemma, then pick the model size to download, or manage models in Settings › Tools › AI › Model Providers › Gemma. Android Studio downloads, verifies and updates the weights itself, so there’s no third-party runtime to install. Check Google’s hardware requirements before choosing a size: 12GB of RAM covers only the smallest models. Because Gemma 4 supports tool calling, the agent can run multi-file refactoring plans locally, which suits projects whose code can’t leave the machine.
How Android skills work
Android skills are modular instructions written by the Android team, following the open agent skills specification, to stop models from writing outdated code for fast-moving APIs. Quail 4 preloads 23 of them, including AGP 9 Upgrade, Android Profiler, Navigation3 and Adaptive. You don’t call them by hand: Android Studio matches your prompt against each skill’s metadata and invokes the relevant ones. You can write custom skills for your team’s workflows, or turn bundled skills off with an IDE-wide toggle in Settings.
Using Android skills outside Android Studio
To give other command-line agents the same knowledge, install the Android CLI and run android skills add --all; the Android CLI guide also covers reserving a real device with Device Streaming. If you prefer to run Claude, Codex or another agent inside the IDE itself, see Bring Your Own Agent in Android Studio; for wireless device testing, see ADB Wi-Fi 2.0.
Other agent changes in Quail 4
When you run parallel chats, the Recent Chats panel shows a spinner while an agent is running tools, a red indicator when it needs your input and a blue badge when a background task is ready. Class, function and file names in replies become links, multi-step tasks end in a single Summary of Changes tab with a diff to review, and reasoning steps collapse into blocks. Google says Quail 4 is the final stable release of the Quail series.
Where Censuus fits
Better tools help you ship; being found brings people to what you ship. Censuus ranks apps by the visits they draw, and adding your app is free through the List my app form. Placement comes from traffic and sponsorship: apps climb on the visits they draw, and a sponsor can pay to rise higher. For on-device AI inside your own app, see Gemini Nano on Android.
Frequently asked questions
How do I use Gemma 4 in Android Studio?
In Android Studio Quail 4, select Gemma in the agent model selector, or open Settings › Tools › AI › Model Providers › Gemma, and choose a model to download. Android Studio handles the download and runs it with a bundled inference engine.
How much RAM do I need for Gemma 4 in Android Studio?
Google says the smallest models run with 12GB of RAM and machines with 32GB or more run best.
Does Gemma 4 in Android Studio work offline?
Yes. It runs locally, your source code doesn’t leave your machine, and there are no token quota limits.
How do I use Android skills in Android Studio?
They’re preloaded in Quail 4 and invoked automatically when your prompt matches a skill. You can also add custom skills or turn bundled ones off in Settings.
Can I use Android skills with other AI agents?
Yes. Install the Android CLI and run android skills add --all to use them with other command-line agents.
Guides for app developers
Put your app in the ranking
Censuus ranks apps by the visits they draw and by sponsorship, no bots. Listing is free; sponsorship raises placement.